key: cord-0552723-f1my60gb
authors: Shoeibi, Afshin; Khodatars, Marjane; Alizadehsani, Roohallah; Ghassemi, Navid; Jafari, Mahboobeh; Moridian, Parisa; Khadem, Ali; Sadeghi, Delaram; Hussain, Sadiq; Zare, Assef; Sani, Zahra Alizadeh; Bazeli, Javad; Khozeimeh, Fahime; Khosravi, Abbas; Nahavandi, Saeid; Acharya, U. Rajendra; Shi, Peng
title: Automated Detection and Forecasting of COVID-19 using Deep Learning Techniques: A Review
date: 2020-07-16
journal: nan
DOI: nan
sha: c4bc53414f5c46ccd356a7d74a9cd8b0cfc8b062
doc_id: 552723
cord_uid: f1my60gb

Coronavirus, or COVID-19, is a hazardous disease that has endangered the health of many people around the world by directly affecting the lungs. COVID-19 is a medium-sized, coated virus with a single-stranded RNA. This virus has one of the largest RNA genomes and is approximately 120 nm. The X-Ray and computed tomography (CT) imaging modalities are widely used to obtain a fast and accurate medical diagnosis. Identifying COVID-19 from these medical images is extremely challenging as it is time-consuming, demanding, and prone to human errors. Hence, artificial intelligence (AI) methodologies can be used to obtain consistent high performance. Among the AI methodologies, deep learning (DL) networks have gained much popularity compared to traditional machine learning (ML) methods. Unlike ML techniques, all stages of feature extraction, feature selection, and classification are accomplished automatically in DL models. In this paper, a complete survey of studies on the application of DL techniques for COVID-19 diagnostic and automated segmentation of lungs is discussed, concentrating on works that used X-Ray and CT images. Additionally, a review of papers on the forecasting of coronavirus prevalence in different parts of the world with DL techniques is presented. Lastly, the challenges faced in the automated detection of COVID-19 using DL techniques and directions for future research are discussed.

The novel COVID-19 virus came to light in December 2019 in Wuhan Province, China, where it originated from animals and quickly spread around the world [1] . The easiest way to transmit COVID-19 is through the air and physical contact, such as hand contact with an infected person [2] . The virus inserts itself into the lung cells through the respiratory system and replicates there, destroying these cells [3] . COVID-19 comprises an RNA and is very difficult to diagnose and treat due to its mutation characteristics [4] . The most common symptoms of COVID-19 include fever, cough, and shortness of breath, dizziness, headache, and muscle aches [5] . The virus is so perilous and can provoke the death of people with weakened immune systems [6] . Infectious disease specialists and physicians around the world are working to discover a treatment for the disease. COVID-19 is currently the leading cause of death for thousands of countries worldwide, including the USA, Spain, Italy, China, the United Kingdom, Iran, and others. Figure 1 exhibits the latest number of infected people worldwide due to COVID-19.

The detection of COVID-19 is crucially significant and vital in its early stages. Various methods have been proposed to diagnose COVID-19, containing a variety of medical imaging techniques, blood tests (CBCs), and PCR. According to the WHO, all diagnoses of corona disease must be confirmed by reverse-transcription polymerase chain reaction (RT-PCR) [7] . However, testing with RT-PCR is highly time-consuming, and this issue is risky for people with COVID-19. Hence, first, medical imaging is carried out for the primary detection of COVID-19, then the RT-PCR test is performed to aid the physicians in making final accurate detection. Two medical imaging techniques, X-ray and CT-scan, are employed to diagnose COVID-19 [8] , [9] .

X-ray modality is the first procedure to diagnose COVID-19, which has the advantage of being inexpensive and low-risk from radiation hazards to human health [10] , [11] . In the Xray method, detecting COVID-19 is a relatively complicated task. In these images, the radiologist must attentively recognize the white spots that contain water and pus, which is very prolonged and problematic. A radiologist or specialist doctor may also mistakenly diagnose other diseases, such as pulmonary tuberculosis, as COVID-19 [12] . The X-ray procedure has a high error rate; hence CT images can be used for more precise detection [14] . Nevertheless, these CT images are far more expensive than X-rays for patients [15] . At the time of CT-scan recording, several slices are provided from each person suspected of COVID-19. The large volume of CT-Scan images calls for a high workload on physicians and radiologists to diagnose COVID- 19. In recent years, applications of artificial intelligence in medicine have led to a variety of studies aiming to diagnose varied diseases, including brain tumors from MR images [16] , [17] , multiple types of brain disorders such from EEG [18] , [19] , breast cancer from mammographic images [20] , [21] and pulmonary diseases such as Covid-19 from X-Ray [22] and CT-Scan [23] . In the last decade, Deep Learning (DL), a branch of machine learning, has changed the expectations in many applications of artificial intelligence in data processing by reaching human-level accuracies [24] in many tasks, including medical image analysis [25] .

In this paper, an overview of COVID-19 diagnostic approaches utilizing DL networks is presented. Section II explains the search strategy, and various DL models developed for COVID-19 detection are described in Section III. Section IV of the DL techniques used for the detection, segmentation, and prediction of COVID-19 patients. Section V discusses the reviewed papers on diagnosis, segmentation, and prediction of COVID-19 patients. Challenges in diagnosing, segmentation, and prediction of COVID-19 patients are provided in Section VI. Finally, the summary and future work are delineated in Section VII.

In this study, valid databases, including IEEE Xplore, Sci-enceDirect, SpringerLink, ACM, and ArXiv, have been used to search for Covid-19 papers. Moreover, a more detailed Google Scholar search is employed. The articles are selected using the keywords COVID-19, Corona Virus, Deep Learning, Segmentation, and Forecasting. The latest selection of papers is done with the mentioned keywords on July 11th, 2020. Figure 2 indicates the number of papers published or indexed by COVID-19 using DL techniques using various databases. 

Traditional machine learning and DL are the two significant branches of AI, but DL is essentially a more advanced version of traditional machine learning (ML). Various DL networks have been extensively used to diagnose the COVID-19 accurately using many public databases.

DL architectures, namely convolutional neural networks (CNNs), recurrent neural networks (RNNs), Autoencoders (AEs), deep belief networks (DBNs), generative adversarial networks (GANs), and hybrid networks such as CNN-RNN and CNN-AE have been developed for automated detection of COVID-19. Figure 3 exhibits the subcategories of DL networks.

Many CADS have been developed using DL methods using X-ray and CT images. Two types of systems: (i) classification and (ii) segmentation using DL methods have been developed. In classification-based CADS, the main objective is to identify COVID-19 patients, which involves the process of extracting features, selecting features, and classifying using deep layers. The second type, CAD, is for the segmentation of X-Ray and CT-Scan images of each infected person with COVID-19. Segmentation implies dividing images into meaningful areas and is of particular notability in medicine. Manual segmentation of medical images takes much time; thus, applying machine learning models is crucially paramount. Among the most important segmentation models, the various types of fuzzy clustering methods [26] , [27] and DL procedures such as U-Net [28] can be expressed. In the CADS, with the segmentation approach, patients' CT-Scan images and their manual segments labeled by doctors are fed to the DL network. Then, during the training process, the DL network is trained on manual segments to segment raw input images. Finally, in deep network output, segmented images are presented with segmentation accuracy. The components of DL-based CADS for COVID-19 detection are shown in Figure 4 . In [33] CT https://github.com/KevinHuRunWen/COVID-19 COVIDx [34] X-ray https://github.com/lindawangg/COVID-Net the following section, we will first mention the important data available for COVID-19. Then, the DL methods exploited in the research are introduced.

Various public databases (X-ray and CT images) available for the detection and prediction of COVID-19 are listed in Table I . Also, the databases used to predict the corona spread in leading countries of the world are shown in Table II .

DL networks developed for classification, segmentation, and prediction of COVID-19 disease will be analyzed in this section. Various applied DL architectures are discussed briefly in the following sections.

1) Classification Models: Various DL methods presented for the automated detection of COVID-19 are discussed in this section. 2D CNN, AlexNet, Visual Geometry Group (VGG) network, GoogLeNet, DenseNet, XceptionNet, Mo-bileNet, SqueezeNet, Inception-ResNet, CapsNet, NasNetmobile, ShuffleNet, EfficientNet, and Generative Adversarial Networks (GAN) have been used for the automated detection of COVID-19 patients using X-ray and CT images. STANDARD 2D-CNN The primary issue in training the deep models is the concern of overfitting that occurs from the gap between the limited number of training samples and a large number of learnable parameters. Convolutional networks try to overcome this by using convolutional layers. CNNs require minimal pre-processing by considering the 2-dimensional (2D) images as input, and hence it is designed to retain and utilize the structural information among neighboring pixels or voxels. A differentiable function is utilized to transform one volume of actions by each layer to the other as it is a sequence of layers structurally. Figure 6 represents the architecture for a usual computer vision job that comprises of three neural layers: convolutional, pooling, and fully connected layers. The convolutional layers are usually combined with the pooling layers, and their output is fed to the fully connected layers [35] . Also, a variety of methods like dropout and batch normalization help these networks to learn better [35] .

ALEXNET As the first famous deep learning network, Alexnet is still the center of attention in many studies. Figure 6 depicts the architecture of AlexNet. In this network, two new perspectives dropout, and local response normalization (LRN) are used to help the network learn better. Dropout is applied in two FC layers employed in the end. On the other hand, LRN, utilized in convolutional layers, can be employed in two different ways: Firstly, applying single channel or feature maps, where the same feature map normalizes depending on the neighborhood values and selects the NN patch. Secondly, LRN can be exploited across the channels or feature maps [36] , [37] .

VGGNET The VGG architecture comprised of a few convolutional layers, each of that utilizes the ReLU activation function. For classification, this network uses a softmax classifier in the final layer of the model. Filter size for convolutional layers is picked equal to 3x3, with a stride of 2 in VGG-E. VGG-11, VGG-16, and VGG-19 are three variants of the VGG-E model that have 11, 16, and 19 layers correspondingly. All variants of VGG-E architecture end with three FC layers. Nevertheless, the numbers of convolution layers are different; VGG-19 contains 16 convolution layers, VGG-16 has 13 convolution layers, and VGG-11 has eight convolution layers. Figure 7 depicts the building block of the VGG network used for COVID-19 detection [36] , [38] .

GOOGLENET Different receptive fields, generated by various kernel sizes, form Inception layers, which are incorporated in this model. Operations generated by these receptive fields records sparse correlation patterns in the novel feature map stack [36] . Figure  8 describes the initial concept of the inception layer. A stack of inception layers is utilized by GoogLeNet to enhance recognition accuracy, as shown in Figure 9 . The difference between the final inception layer and the nave inception layer is the inclusion of 1x1 convolution kernels, which performs a dimensionality reduction, consequently reducing the computational cost. Another idea in GooGLeNet is the gradient injection, which aims to overcome the gradient vanishing problem. GoogLeNet comprises of a total of 22 layers that is greater than any previous network. However, GoogLeNet uses much fewer parameters compared to its predecessors VGG or AlexNet [36] , [39] .

RESNET The Residual Network (ResNet) is created with various numbers of layers; 1202,152, 101, 50, and 34. ResNet50 is one of the popular variants containing 49 convolution layers and 1 FC layer at the end of it. The total number of MACs and weights are 3.9M and 25.5M, respectively [36] , [40] , [41] . Figure 10 shows a typical ResNet architecture used for COVID-19 detection.

DENSENET The DenseNet that comprised of densely connected CNN layers, in a dense block [36] , [43] with the outputs of each layer are connected with all descendant layers. Due to the dense connectivity between the layers, it is termed as DenseNet. Network parameters are reduced dramatically by efficient utilization of feature reuse. DenseNet comprises of various transition blocks and dense blocks that are situated in between two adjacent dense blocks. Figure 11 describes the conceptual diagram of a dense block [36] .

XCEPTIONNET The Xception architecture is based on Inception V3. The Xception architecture is a linear stack of depth-wise separable convolution layers with residual connections. The network entails 36 layers of convolution organized in 14 modules, all of which contain linear residual connections around them, except for the first and last modules. Utilizing residual connections in Xception architecture will lead to faster convergence and superior final performance [44] . MOBILENET The novel model called MobileNets is aimed to be used [36] .

in mobile and embedded machine vision applications [45] .

The main layers exploited in this architecture are known as depthwise separable convolution. The depthwise separable convolution consists of two layers: depthwise convolution, and pointwise convolution. The depthwise convolution is used to apply a single filter to each input channel (input depth). Then the pointwise convolution, simple 1x1 convolution, is employed to create a linear combination of depthwise layer outputs. After both of these layers, a ReLU and batch normalization (BN) are placed ( Figure 12 ) [46] . SQUEEZENET Using three architectural design strategies, this structure has introduced a fire module [47] . The module consists of a squeeze-convolution layer that has only 1x1 filters, followed by an expansion layer that has a set of 1x1 and 3x3 convolution filters. With slight modification and optimization in the original SqueezeNet architecture, two other architectures, SqueezeNet with simple bypass and SqueezeNet with complex bypass, were also demonstrated. By benchmarking the SqueezeNet architecture on the ImageNet database and comparing its results with the AlexNet architecture, it was found that the SqueezeNet accuracy is at the AlexNet level, but the 50X has a lower parameter, and the model size is less than 0.5MB [47] .

INCEPTION-RESNET In 2016, Szegedy et al. proposed the idea of integrating Inception architecture with residual connections, meaning that the filter concatenation stage in Inception architecture would be replaced by residual connections [41] . The idea led to a new architecture called Inception-ResNet, which has two versions, V1 and V2. At the same time, another architecture named Inception V4 was introduced, in which reduction blocks were exploited. These blocks are modified versions of Inception modules. Finally, it was perceived that introducing residual connections to Inception architecture significantly improves training speed. Also, both of these architectures have outperformed than previous architects [41] .

CAPSNET With advancements of CNNs and new structures, they have reached high accuracies on many tasks. However, one of the deficiencies of CNN models is when they face samples drown from a dataset with a different orientation than a training dataset. To address this, CapsNet was proposed. The central idea behind this network is to create a network that implicitly performs an operation similar to inverse graphics; i.e., it tries to find graphical shapes in an image. The building block of this structure is capsules, which try to determine whether an object is presented at a given location and find its instantiation. With the help of these capsules, CapsNet performs better than its prior models in many tasks and specifically in cases where two classes have a considerable overlap [48] .

NASNET-MOBILE In a try to find a method to learn architecture directly from data instead of handly designing it [49] , NASNet was created. First, the best normal convolution cell and the reduction convolution cell are found in NASNet's search space by applying the RNN controller, on a small dataset. The RNN controller then stacks multiple copies of these cells with various parameters to acquire NASNet architecture. A new regularization technique called ScheduledDropPath has also been stated, which dramatically meliorate the generalizability of NASNet models. The architectures obtained on the COCO object detection database have also been evaluated in all cases, showing that the architectures could achieve state-of-the-art performance [49] .

SHUFFLENET ShuffleNet is specially designed for mobile devices with minimal computing power [50] , [51] . This architecture employs two operations, pointwise group convolution and channel shuffle, to maintain the network's accuracy while reducing the computational cost. ShuffleNet architecture embraces a convolution layer, two pooling layers, a stack of ShuffleNet units that are structured in three stages, and finally, an FC layer. Although ShuffleNet is designed for small models, it still surpasses MobileNet (lower computation cost, higher training speed). Figure 13 shows a general form of ShuffleNet used for COVID-19 detection.

EFFICIENTNET The fundamental building block in EfficientNet was to overcome the MBConv mobile bottleneck. The EfficientNet architecture developed using the compound scaling method, which led to the EfficientNet-B0 to B7. It is found that this architecture has fewer parameters of 8.4x and a faster running time of 6.1x [53] .

2) Generative Adversarial Networks (GAN): A primary problem in training deep models is limits in dataset size. Using generative models for data augmentation is one solution to this issue. Due to the high quality of generated data, GANs have attracted attention in the medical imaging community [54] . The basic idea in training a GAN is a simple minimax game, in which one network tries to distinguish between real data and generates one, and the other tries to create data undistinguishable by the first network [55] , therefor creating images similar to real data. Figure 14 shows a simple gan architecture.

3) Segmentation Models: In this section, various DL models developed to segment the lung region to detect the COVID-19 patients accurately are discussed. FCN network, SegNet, U-Net, and Res2Net DL models are widely used for the segmentation of lungs and are briefly discussed below.

FULLY CONVOLUTIONAL NETWORK (FCN) In this model, popular networks have transformed entirely convolutional models by replacing FC layers with convolution layers to capture output as a local map. These maps are upsampled using the introduced method, which uses a backward convolution with stride size f, capable of learning. At the end of the network, there is a 1x1 convolution layer that yields the corresponding pixel label as the output. The exiting stride in the deconvolution stage constraints the output detail quantity of this layer. To address this issue and enhance the quality of the result, several skip connections have been added to the network from the lower layers to the end layer [56] . Figure  15 shows a general form of FCN used for the segmentation of lung in COVID-19 patients.

SEGNET Generally, in segmentation techniques, a network created for classification is chosen, and the FC layers of that network are removed; the resulting network is called the encoder network. Then a decoder is created to transform these low-resolution maps to the original resolution. In SegNet [57] , the decoder is created such that for each down-sampling layer in the encoding section, an up-sampling layer is positioned in the decoder. These layers, unlike the deconvolution layers of FCN networks, are not capable of learning, and the values are placed at the locations from which the corresponding max-pooling layer is extracted, and the rest of the output cells become zero. Figure 16 shows a general form of SegNet used for the segmentation of lung in COVID-19patients.

U-NET The U-Net network [59] , like SegNet, consists of the identical numbers of pooling and up-sampling layers, but the network utilizes trainable deconvolution layers. Also, in this network, there is a corresponding skip connection between the up-sampling and down-sampling layers. Figure 17 shows RES2NET

In the Res2Net module, after the 1*1 convolution, the feature maps are divided into several subsets, then passed through a set of 3*3 filters. Their outputs are concatenated together and then go through the 1*1 convolution [61] . The set of this process is residually structured. For that reason, it is named as Res2Net module. This module introduces a new control parameter called the scale dimension (the number of feature groups in the Res2Net block); with an increase of scale, features with richer receptive field sizes are learned by the model. The Res2Net module is capable of integrating with modern modules such as cardinality dimension and squeeze and excitation (SE). It can also be integrated easily with stateof-the-art models, such as ResNet, ResNeXt, DLA, Big-Little Net, which are called Res2Net, Res2NeXt, Res2Net-DLA, and bLRes2Net-50, respectively [61] .

A feed-forward neural network is extended to create RNN, aiming to capture the long term dependencies and features from the sequential and time-series data. The most commonly used RNN is the long-short term memory (LSTM), which composed of a memory cell Ct, a forget cell ft, the input gate it, and output gate ot ( figure 18(a) ). These gates make the decision that which information needs to be remembered or discarded from the memory cell and also organizes the activation signals from different sources.

LSTM decides whether to keep or remove the memory by using these gates, unlike vanilla RNN, LSTM can preserve the potential long term dependencies of a feature, which is learned from the input sequential data. One LSTM variant is Gated Recurrent Unit (GRU) [62] , which integrates the forget and input gates into a single update gate and combines the memory cell state and the hidden state into one state ( figure 18(b) ). Update gate makes a decision on the amount of information to be added or discarded, and the reset gate decides on how much earlier information is to be forgotten. This technique makes GRU simpler than LSTM. AUTOENCODERS (AES) Autoencoder (AE) is a neural network method with competent data encoding and decoding strategies used for unsupervised feature learning [35] . The primary purpose of AE is usually to learn and representation of data (encoding), as well as dimensionality reduction of data, fusion, compression, and many more [63] , [64] . The AE models comprised of two phases: encoder and decoder. The input samples are mapped typically to a lower-dimensional space with beneficial feature representation in the encoding part. Reverse processing is applied in the decoding phase to revert data to its original space, trying to create data from lower space representation. Figure 19 depicts the conceptual diagram of AE with encoding and decoding phases. 

The main focus of this work is to select the best DL models employed to detect, segment the lungs, and predict the COVID-19 patients using DL techniques. The summary of works done on classification, segmentation, and prediction are presented in Tables III, IV , and V, respectively. Figure  20 depicts the total number of investigations conducted in the field of classification, segmentation, and prediction of COVID-19 using DL models. It can be noted from the figure that most works have been done on the detection of COVID-19 patients, and the least works are done on the forecasting due to shortage of available public databases.

The X-ray and CT images have been used to develop classification and segmentation DL models. Figure 21 shows the total number of times each modality is used in reviewed studies. It can be observed that most of the researchers have used X-ray images. This may be due to cheaper registration fees, and the fact that slice selection is not needed. Also, very few researches have used combined modalities of X-ray and CT images due to the absence of such a comprehensive database.

Various DL models developed for the automated detection of COVID-19 patients is shown in Figure 22 . It can be noted from the figure that; different types of convolutional networks have been commonly used. Also, for the automated segmentation of lungs, various types of U-Net are more common.

Nowadays, a variety of toolboxes have been used to implement DL models. The number of toolboxes used for automated detection of COVID-19 by researches is shown in Figure 23 . It can be noted that Keras toolbox is the most widely used for automated detection of COVID-19 patients; this is due to its simplicity and also availability of pre-trained models in this library, which are widely used by researchers.

The last part of this study is devoted to the classification algorithms developed using DL architectures. The softmax is most employed for automated detection of COVID-19 patients (Tables III to V) . Figure 24 shows the number of 

With the rapid growth and spread of COVID-19 globally, researchers have confronted many serious challenges in designing and implementing CADS to diagnose and predict the disease. The most significant challenges associated with COVID-19 are data availability, DL networks architecture fixing, and hardware resources. Lack of availability of a huge public database comprising X-ray and CT images is the first challenge. Due to the limited number of patient data, many researchers have used pre-trained networks such as GoogLeNet and AlexNet. Nevertheless, the number of studies conducted on forecasting is limited as it requires a vast database.

One of the problems of employing pre-trained networks is that these models are often trained on the ImageNet database, which is entirely different from medical images. Hence, implementing efficient CADS to accurately and swiftly diagnose COVID-19 from X-Ray or CT images is still a challenging work. Physicians are not just convinced with X-ray or CT-scan images of patients to accurately diagnose COVID-19; they may use both modalities simultaneously. However, complete and comprehensive databases of the X-ray and CT-scan hybrid modalities for CADS research and implementation have not been provided for researchers' in the machine learning scope. For this reason, researchers combine different X-ray and CT-Scan datasets from various datasets, which may disrupt network training. Yielding the combined X-Ray and CT-Scan datasets pave the path to help quickly identify COVID-19 alongside DL networks. The third challenge in the data section is the non-reporting of phenotypic information, such as age and gender. The utilization of this information can amend and enhance the performance of DL algorithms. Table IV summarizes the DL-based segmentation algorithms aimed at identifying areas suspected of corona in the X-ray and CT-scan images. One of the obstacles with databases is the absence of manual or ground truths for COVID-19 image segmentation areas. Therefore, many researchers have delineated these areas with the help of radiologists and trained the models such as U-Net, which is time-consuming. Consequently, the presence of dedicated databases of segmented images will help to get the best performing model. Also, it becomes easy to compare the performances with other authors who have worked on the same images.

In order to predict the prevalence of corona using DL methods, the nature of the COVID-19 is still relatively unknown, and the probability of mutation is a big issue. Therefore, to predict the prevalence of the disease, many factors like the average age of the society, policies to impede the spread of the disease by countries, climatic conditions, and infection of neighbor/friend/family member.

Lack of access to appropriate hardware resources is another challenge. Implementing DL architectures in CADS for corona diagnosis demands strong hardware resources, which unfortunately is not ordinarily accessible for many researchers. Although tools such as Google Colab have partially obviated this problem, employing these tools in real medical applications is still challenging. For this reason, in most studies, researchers have not provided practical CADS systems such as web or Windows software to detect COVID-19.

COVID-19 is an emerging pandemic disease that, in a short period of time, can severely endanger the health of many people throughout the world. It directly affects the lung cells, and if not accurately diagnosed early, can cause irreversible damage, including death. The disease is accurately detected by the specialists using X-ray or CT images together with PCR results. The PCR results indicate the type of lung disease, such as pulmonary tuberculosis, instead of COVID-19. In this study, a comprehensive review of the accomplished studies of COVID-19 diagnosis was carried out using DL networks. The public databases available to diagnose and predict COVID-19 are presented. The state-of-art DL techniques employed for the diagnosis, segmentation, and forecasting of the spread of COVID-19 are presented in Tables III, IV , V, respectively. One of the challenges to develop a robust and accurate COVID 19 diagnosis system is the availability of an extensive public database. We strongly feel that, with more public databases, better DL models can be developed by researchers to detect and predict the COVID19 accurately. Hence, this will help to develop the best performing model. We feel that data fusion models can help to improve the performance of diagnosis and prediction. The features extracted from the ML and DL models can be fused to develop an accurate model. Also, developing the accurate segmentation model is challenging as it involves delineating the lungs by the experts. Having an accurate ground truth is another challenge.

This section discusses the evaluation metrics used in different studies. In these metrics, True positive (TP) is the correct classification ratio of the positive class, False positive (FP) is the incorrect prediction ratio of the positives, True negative (TN) is the correct classification ratio of the negative class, and finally False negative (FN) is the incorrect prediction ratio of the negatives [189] .

The receiver operating characteristic curve (ROC-curve) illustrates the performance of the proposed model at all classification thresholds. It is the graph of true positive rate vs. false positive rate (TPR vs. FPR).

T P R = T P T P + F N (1)

B. Area under the ROC Curve (AUC)

AUC depicts the area under the ROC-curve incorporated from (0, 0) to (1, 1) . It presents the cumulative measure of all possible classification thresholds. AUC has a range from 0 to 1. A 100% wrong classification will have AUC value 0.0, while a 100% correct classified version will have the AUC value 1.0. It has two folded benefits; first, it is scale-invariant, meaning that entailed value shows how well the model is predicted rather than examining the absolute values. The second benefit is that it is classification-threshold invariant, as it will validate the models performance irrespective of the threshold being chosen.

Accuracy verifies that how many samples are correctly classified.

Accuracy(Acc) = T P + T N T P + T N + F P + F N (3)

It is the rate of recognition of negative samples correctly.

Specif icity(Spe) = T N T N + F P (4)

It is the rate of recognition of positive samples correctly.

Sensitivity(Sen) = T P T P + F N (5)

It calculates how precise the model performs by examining the correct true positives from the predicted ones.

P recision(P re) = T P T P + F P

G. F1-Score F1-score is the function of sensitivity and precision, which tries to find a balance between sensitivity and precision. F 1 − Score = 2 * P re * Sen P re + Sen (7)

It is a measure of effectiveness of the classification technique.

Avg Acc = n T P +T N T P +T N +F P +F N n (8) where n is the total number of outputs of the system.

It is a measure of similarity rate between two sample groups.

Jac idx = T P T P + F P + F N (9)

It is a statistical measure of similarity rate between two sample groups.

Dice cof = 2 * T P 2 * T P + F P + F N

A familial cluster of pneumonia associated with the 2019 novel coronavirus indicating person-to-person transmission: a study of a family cluster

Transmission routes of 2019-ncov and controls in dental practice

Pulmonary pathology of early phase 2019 novel coronavirus (covid-19) pneumonia in two patients with lung cancer

Structure of the full sarscov-2 rna genome in infected cells

Coronavirus disease 2019 (covid-19): a guide for uk gps

Covid-19 and italy: what next

World health organization declares global emergency: A review of the

Automatic x-ray covid-19 lung image classification system based on multi-level thresholding and support vector machine

Development of a machine-learning system to classify lung ct scan images into normal/covid-19 class

High discordance of chest x-ray and computed tomography for detection of pulmonary opacities in ed patients: implications for diagnosing pneumonia

The role of chest imaging in patient management during the covid-19 pandemic: a multinational consensus statement from the fleischner society

Covid-19 in diabetic patients: related risks and specifics of management

Risk factors prediction, clinical outcomes, and mortality of covid-19 patients

Added value of ultra-low-dose computed tomography, dose equivalent to chest x-ray radiography, for diagnosing chest pathology

Deep neural network with generative adversarial networks pre-training for brain tumor classification based on mr images

Convolutional neural networks for multi-class brain disease detection using mri images

Epileptic seizures detection in eeg signals using tqwt and ensemble learning

Deep convolutional neural network for the automated detection and diagnosis of seizure using eeg signals

A hierarchical classification method for breast tumor detection

Deep learning modeling using normal mammograms for predicting breast cancer risk

Covid-classifier: An automated machine learning model to assist in the diagnosis of covid-19 infection in chest x-ray images

Artificial intelligence-enabled rapid diagnosis of patients with covid-19

Dermatologist-level classification of skin cancer with deep neural networks

Deep learning in radiology: An overview of the concepts and a survey of the state of the art with focus on mri

Semisupervised fuzzy clustering methods for x-ray image segmentation

An efficient brain mr images segmentation hardware using kernel fuzzy c-means

U-net: Convolutional networks for biomedical image segmentation

Covid-ct-dataset: a ct scan dataset about covid-19

Pocovid-net: Automatic detection of covid-19 from a new lung ultrasound imaging dataset (pocus)

A large-scale covid-19 twitter chatter dataset for open scientific research-an international collaboration

Automated diagnosis of covid-19 using deep learning and data augmentation on chest ct

Covid-net: A tailored deep convolutional neural network design for detection of covid-19 cases from chest radiography images

A survey on deep learning for neuroimaging-based brain disorder analysis

The history began from alexnet: A comprehensive survey on deep learning approaches

Imagenet classification with deep convolutional neural networks

Very deep convolutional networks for large-scale image recognition

Going deeper with convolutions

Identity mappings in deep residual networks

Inception-v4, inception-resnet and the impact of residual connections on learning

Deepcovidexplainer: Explainable covid-19 predictions based on chest x-ray images

Densely connected convolutional networks

Xception: Deep learning with depthwise separable convolutions

Mobilenets: Efficient convolutional neural networks for mobile vision applications

Mo-bilenetv2: Inverted residuals and linear bottlenecks

Squeezenet: Alexnet-level accuracy with 50x fewer parameters and¡ 0.5 mb model size

Dynamic routing between capsules

Learning transferable architectures for scalable image recognition

Shufflenet: An extremely efficient convolutional neural network for mobile devices

Shufflenet v2: Practical guidelines for efficient cnn architecture design

Automated diagnosis of covid-19 using deep learning and data augmentation on chest ct

Efficientnet: Rethinking model scaling for convolutional neural networks

Generative adversarial network in medical imaging: A review

Generative adversarial nets

Fully convolutional networks for semantic segmentation

Segnet: A deep convolutional encoder-decoder architecture for image segmentation

Covid-19 chest ct image segmentationa deep convolutional neural network solution

U-net: Convolutional networks for biomedical image segmentation

An automatic covid-19 ct segmentation network using spatial and channel attention mechanism

Res2net: A new multi-scale backbone architecture

Empirical evaluation of gated recurrent neural networks on sequence modeling

Autoencoders, unsupervised learning, and deep architectures

Wasserstein auto-encoders

Covid-resnet: A deep learning framework for screening of covid19 from radiographs

Automatic detection of coronavirus disease (covid-19) using x-ray images and deep convolutional neural networks

Diagnosing covid-19 pneumonia from x-ray and ct images using deep learning and transfer learning algorithms

Covidx-net: A framework of deep learning classifiers to diagnose covid-19 in x-ray images

Covid-19 screening on chest x-ray images using deep learning based anomaly detection

Covid-net: A tailored deep convolutional neural network design for detection of covid-19 cases from chest x-ray images

Covidiagnosis-net: Deep bayes-squeezenet based diagnostic of the coronavirus disease 2019 (covid-19) from x-ray images

Estimating uncertainty and interpretability in deep learning for coronavirus (covid-19) detection

Covid-19 detection using deep learning models to exploit social mimic optimization and structured chest x-ray images using fuzzy color and stacking approaches

Deep learning-based multi-view fusion model for screening 2019 novel coronavirus pneumonia: a multicentre study

Automated detection of covid-19 cases using deep neural networks with x-ray images

Application of deep learning technique to manage covid-19 in routine clinical practice using ct images: Results of 10 convolutional neural networks

Classification of covid-19 in chest x-ray images using detrac deep convolutional neural network

Covid-19 identification in chest x-ray images on flat and hierarchical classification scenarios

Detection of coronavirus (covid-19) associated pneumonia based on generative adversarial networks and a fine-tuned deep transfer learning model using chest x-ray dataset

Coronavirus (covid-19) classification using deep features fusion and ranking technique

Covid-caps: A capsule network-based framework for identification of covid-19 cases from x-ray images

Coronet: A deep neural network for detection and diagnosis of covid-19 from chest x-ray images

Deep learning on chest x-ray images to detect and evaluate pneumonia cases at the era of covid-19

Towards an efficient deep learning model for covid-19 patterns detection in x-ray images

Radiologistlevel covid-19 detection using ct scans with detail-oriented capsule networks

Gsa-densenet121-covid-19: a hybrid deep learning architecture for the diagnosis of covid-19 disease based on gravitational search optimization algorithm

Deepcovid: Predicting covid-19 from chest x-ray images using deep transfer learning

Deep convolutional approaches for the analysis of covid-19 using chest x-ray images from portable devices

Robust screening of covid-19 from chest x-ray via discriminative cost-sensitive learning

Automated diagnosis of covid-19 with limited posteroanterior chest x-ray images using fine-tuned deep neural networks

Deep learning for screening covid-19 using chest x-ray images

A new modified deep convolutional neural network for detecting covid-19 from x-ray images

Covid-da: Deep domain adaptation from typical pneumonia to covid-19

Fast and accurate detection of covid-19-related pneumonia from chest x-ray images with novel deep learning model

Multi-channel transfer learning of chest x-ray images for screening of covid-19

Deep learning enables accurate diagnosis of novel coronavirus (covid-19) with ct images

Adaptive feature selection guided deep forest for covid-19 classification with chest ct

Improving coronavirus (covid-19) diagnosis using deep transfer learning

Classification of coronavirus images using shrunken features

A deep learning algorithm using ct images to screen for corona virus disease

Training deep learning algorithms with weakly labeled pneumonia chest x-ray data for covid-19 detection

Covidctnet: An open-source deep learning approach to identify covid-19 using ct image

Automatic detection of covid-19 infection from chest x-ray using deep learning

Coronet: A deep network architecture for semi-supervised task-based identification of covid-19 from chest x-ray images

Sample-efficient deep learning for covid-19 diagnosis based on ct scans

Dual-sampling attention network for diagnosis of covid-19 from community acquired pneumonia

Covid-densenet: A deep learning architecture to detect covid-19 from chest radiology images

Deep learning for detecting corona virus disease 2019 (covid-19) on high-resolution computed tomography: a pilot study

Detection of coronavirus disease (covid-19) based on deep features

Deep learning system to screen coronavirus disease 2019 pneumonia

Extracting possibly representative covid-19 biomarkers from x-ray images with deep learning approach and image data related to pulmonary diseases

Covidgan: Data augmentation using auxiliary classifier gan for improved covid-19 detection

Using x-ray images and deep learning for automated detection of coronavirus disease

Within the lack of chest covid-19 x-ray dataset: A novel detection model based on gan and deep transfer learning

Classification of covid-19 coronavirus, pneumonia and healthy lungs in ct scans using q-deformed entropy and deep learning features

Finding covid-19 from chest x-rays using deep learning on a small dataset

Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks

A deep convolutional neural network for covid-19 detection using chest x-rays

Predicting covid-19 pneumonia severity on chest x-ray with deep learning

Can ai help in screening viral and covid-19 pneumonia

A deep transfer learning model with classical data augmentation and cgan to detect covid-19 from chest ct radiography digital images

A modified deep convolutional neural network for detecting covid-19 and pneumonia from chest x-ray images based on the concatenation of xception and resnet50v2

Classification of x-ray images for detecting covid-19 using deep transfer learning

Intra-model variability in covid-19 classification using chest x-ray images

A study of the neutrosophic set significance on deep transfer learning models: An experimental case on a limited covid-19 chest x-ray dataset

Automatic detection of coronavirus disease (covid-19) in x-ray and ct images: A machine learning-based approach

Deep learning covid-19 detection bias: accuracy through artificial intelligence

Any unique image biomarkers associated with covid-19?

End-to-end covid-19 screening with 3d deep learning on chest computed tomography

Deep transfer learning based classification model for covid-19 disease

Application of deep learning for fast detection of covid-19 in x-rays using ncovnet

Fast deep learning computeraided diagnosis against the novel covid-19 pandemic from digital chest x-ray images

Explainable deep learning for pulmonary disease and coronavirus covid-19 detection from x-rays

Covid-mobilexpert: On-device covid-19 screening using snapshots of chest x-ray

Classification of covid-19 patients from chest ct images using multi-objective differential evolutionbased convolutional neural networks

Covxnet: A multidilation convolutional neural network for automatic covid-19 and other pneumonia detection from chest x-ray images with transferable multireceptive feature optimization

End-to-end automatic differentiation of the coronavirus disease 2019 (covid-19) from viral pneumonia based on chest ct

Truncated inception net: Covid-19 outbreak screening using chest x-rays

A novel and reliable deep learning web-based tool to detect covid-19 infection form chest ctscan

A deep convolutional neural network for covid-19 detection using chest x-rays

Covidlite: A depth-wise separable deep neural network with white balance and clahe for detection of covid-19

Exploration of interpretability techniques for deep covid-19 classification using chest x-ray images

Deep convolutional neural networks to diagnose covid-19 and other pneumonia diseases from posteroanterior chest x-rays

Vulnerability of deep neural networks for detecting covid-19 cases from chest x-ray images to universal adversarial attacks

Detection of coronavirus disease (covid-19) based on deep features and support vector machine

Covid-cxnet: Detecting covid-19 in frontal chest x-ray images using deep learning

A noise-robust framework for automatic segmentation of covid-19 pneumonia lesions from ct images

Rapid ai development cycle for the coronavirus (covid-19) pandemic: Initial results for automated detection & patient monitoring using deep learning ct image analysis

Lung infection quantification of covid-19 in ct images with deep learning

Covid-ct-dataset: a ct scan dataset about covid-19

Covid mtnet: Covid-19 detection with multi-task deep learning approaches

Residual attention u-net for automated multi-class segmentation of covid-19 chest ct images

Coronavirus detection and analysis on chest ct with deep learning

Jcs: An explainable covid-19 diagnosis system by joint classification and segmentation

An automatic covid-19 ct segmentation based on u-net with attention mechanism

Unveiling covid-19 from chest x-ray with deep learning: a hurdles race with small data

Relational modeling for robust and efficient pulmonary lobe segmentation in ct scans

Deep learning covid-19 features on cxr using limited training data sets

Weakly supervised deep learning for covid-19 infection detection and classification from ct images

Miniseg: An extremely minimum network for efficient covid-19 segmentation

Iteratively pruned deep learning ensembles for covid-19 detection in chest x-rays

Inf-net: Automatic covid-19 lung infection segmentation from ct images

A cascade network for detecting covid-19 using chest x-rays

Deep learning models for covid-19 infected area segmentation in ct images

Multi-task deep learning based ct imaging analysis for covid-19: Classification and segmentation

Deep learning-based detection for covid-19 from chest ct using weak label

Ai-assisted ct imaging analysis for covid-19 screening: Building and deploying a medical ai system in four weeks

Prior-attention residual learning for more discriminative covid-19 screening in ct images

Deep learning for classification and localization of covid-19 markers in point-of-care lung ultrasound

Artificial intelligence distinguishes covid-19 from community acquired pneumonia on chest ct

A fully automatic deep learning system for covid-19 diagnostic and prognostic analysis

Deep learning system to screen coronavirus disease 2019 pneumonia

Deep learning to estimate the physical proportion of infected region of lung for covid-19 pneumonia with ct image set

Quantification of tomographic patterns associated with covid-19 from chest ct

Accurate screening of covid-19 using attention based deep 3d multiple instance learning

Ct-based covid-19 triage: Deep multitask learning improves joint identification and severity quantification

Time series forecasting of covid-19 transmission in canada using lstm networks

Multiple-input deep convolutional neural network model for short-term photovoltaic power forecasting

Predicting covid-19 incidence through analysis of google trends data in iran: data mining and deep learning pilot study

Prediction for the spread of covid-19 in india and effectiveness of preventive measures

Deep ensemble learning method to forecast covid-19 outbreak

Artificial intelligence forecasting of covid-19 in china

Covid-19 infection forecasting based on deep learning in iran

Neural network based country wise risk prediction of covid-19

Predicting the epidemic curve of the coronavirus (sars-cov-2) disease (covid-19) using artificial intelligence

An improved method of covid-19 case fitting and prediction based on lstm

Variational-lstm autoencoder to forecast the spread of coronavirus across the globe

Prediction and analysis of covid-19 positive cases using deep learning models: A descriptive case study of india

Cancer diagnosis using deep learning: a bibliographic review